Despite the significant progress in shrimp farming and aquaculture, there is a tendency to overlook the crucial aspect of managing pond wastewater, which poses environmental risks due to untreated effluent release and the associated threat of water pollution. The quality of water plays a vital role in shrimp cultivation outcomes, and the conventional methods for assessing water parameters involve labor-intensive field measurements, introducing time and spatial limitations to the evaluation process. To address these challenges, our proposed solution involves leveraging Sentinel-3 data and utilizing machine learning algorithms for predicting diverse water quality parameters.
We incorporated two types of in situ data sources: in-house and external data. The in-house dataset comprises 23 shrimp ponds from our collection situated along the coastal lines of Java and Sulawesi in Indonesia. For the external data, we included 22 coastal sites sourced from the European Environment Agency (EEA). We used spatial and temporal matching for the in situ data and its corresponding satellite data. The modelling techniques included Support Vector Regression (SVR), Random Forest Regression (RF), XGBoost Regression (XGB), and Stacking Regression (Stack). Table 1 provides a comprehensive overview of the model performance results for parameters such as Ammonium (NH4), Dissolved Oxygen (DO), Nitrate (NO3), Nitrite (NO2), Oxygen Saturation (OS), pH, Salinity (Sal), Total Nitrogen (TN), Total Phosphorus (TP), and Water Temperature (WT).
XGBoost consistently emerged as the superior model across six water parameters, demonstrating robust predictive accuracy for NH4, DO, OS, Sal, and WT. Despite these successes, challenges were observed, notably in achieving optimal performance for parameters such as NO2 and TP. This research underscores the potential of integrating satellite data and machine learning for effective effluent monitoring via water quality prediction in aquaculture, contributing valuable insights for sustainable shrimp farming practices.